GCP Compute Engine — Easy Notes

1. Compute options in Google Cloud

Google Cloud provides different ways to run applications:

Compute Engine → Virtual Machines (VMs)
Google Kubernetes Engine (GKE) → Containers/Kubernetes
Cloud Run → Serverless containers
Cloud Functions → Event-driven serverless
Cloud TPU → Specialized hardware for Machine Learning

This section focuses mainly on Compute Engine.

2. What is Compute Engine?

Compute Engine = Virtual Machines in Google Cloud.

Think of it like:

Your own server running inside Google Cloud.

You get:

VM
Operating System
CPU/vCPU
RAM
Disk
Networking
IP address
Firewall configuration

You have significant control over how the VM operates.

Example

Suppose your company has an old Java application:


Java Application
       ↓
Linux Server
       ↓
CPU + RAM + Disk

Instead of buying a physical server, you can move that workload to:


Google Cloud
     ↓
Compute Engine VM
     ↓
Linux
     ↓
Java Application

This is why Compute Engine is useful for migrating traditional on-premises applications to the cloud.

3. Compute Engine = IaaS

Compute Engine is primarily:

IaaS = Infrastructure as a Service

You manage more of the infrastructure compared with serverless services.


Compute Engine
      ↓
You manage
 ├── OS
 ├── Applications
 ├── Configuration
 ├── Patching
 └── Scaling rules

Google manages the underlying physical infrastructure.

Easy way to remember

IaaS → Google gives you infrastructure; you manage the VM.

4. Main use case

Compute Engine is best for generic workloads, especially applications designed to run on traditional servers.

Examples:

Enterprise Java applications
Web servers
Databases
Legacy applications
Applications migrated from on-premises
Custom software requiring OS-level control

Why is it portable?

If an application already works on a traditional Linux/Windows server, moving it to a Compute Engine VM can be relatively straightforward.

5. Compute Engine machine configuration

You can choose the resources for your VM.

CPU + Memory

You can use:

Predefined machine types

Custom machine types

Custom machine type means you can choose the amount of:


CPU
+
RAM

according to your requirements.

6. vCPU — Important

A vCPU = virtual CPU.

In Compute Engine, each vCPU is implemented as a hardware hyper-thread on the underlying CPU platform.

Think:


Physical CPU
     ↓
Hyper-thread
     ↓
vCPU
     ↓
Your VM

Don't confuse:

Physical CPU core
Hardware thread
vCPU

For GCP VM sizing, you generally select the number of vCPUs.

7. CPU affects network performance

Your choice of CPU/vCPU can also affect network throughput.

The course gives the rule:

Network throughput can scale at approximately 2 Gbps per CPU core, with exceptions for certain smaller instances.

For example, the course mentions:

2 or 4 CPU instances → up to 10 Gbps
A C3 instance with 176 vCPUs → theoretical maximum around 200 Gbps

Important

Don't memorize these numbers as universal rules forever because network limits depend on the machine series and configuration.

For exams/interviews, remember:

More powerful VM configurations can provide higher network throughput.

8. Disk options

Compute Engine provides different disk choices.

The important ones here are:

1. Standard Persistent Disk

Uses traditional HDD-style storage.

Advantages:

Lower cost
Good when very high disk performance isn't required

Think:

Capacity for money

2. SSD Persistent Disk

Uses flash-based SSD storage.

Advantages:

Faster
Higher IOPS
Lower latency than standard HDD-based storage

Think:

Performance for money

3. Local SSD

Local SSD is physically attached to the host hardware.

Therefore:


VM
 ↓
Local SSD
 ↓
Physical host

This provides:

Very high throughput
Very low latency

But there is an important disadvantage:

Local SSD data is temporary/ephemeral.

The data doesn't survive certain VM lifecycle events, particularly when the VM is stopped or deleted.

So don't use local SSD as your primary permanent storage for important data.

9. Persistent Disk vs Local SSD

Feature	Persistent Disk	Local SSD
Storage type	Persistent	Temporary/ephemeral
Performance	Good	Very high
Latency	Higher	Very low
Data survives VM lifecycle	Generally yes	No
Typical use	Application/data storage	Temporary/high-speed data

Easy memory trick

Persistent = Permanent

Local SSD = Fast but temporary

10. SSD vs Standard Disk

The choice is mainly:

Performance vs Cost

Standard HDD


Cheaper
   ↓
More capacity per money
   ↓
Lower performance

SSD


More expensive
   ↓
Better IOPS
   ↓
Better performance

IOPS = Input/Output Operations Per Second

If your application performs many disk operations, SSD is generally preferable.

11. Disk performance and size

For persistent disks, performance can scale with the amount of disk capacity provisioned.

So disk size isn't only about:

"How much data can I store?"

It can also influence:

"How much disk performance can I get?"

12. Operating Systems

Compute Engine supports different operating systems, including:

Linux
Windows

You can therefore run a mixed environment:


VM 1 → Linux
VM 2 → Linux
VM 3 → Windows
VM 4 → Windows

This is useful for enterprise environments with different application requirements.

13. Networking in Compute Engine

Compute Engine VMs integrate with VPC networking.

You can configure:

Network interfaces
IP addresses
Firewall rules
Network tags
Load balancing

Example:


Internet
   ↓
Load Balancer
   ↓
VPC
   ↓
Compute Engine VMs
 ┌─────┬─────┬─────┐
 VM1   VM2   VM3

14. Load Balancing

Compute Engine can work with:

Application Load Balancer

Used mainly for application/HTTP(S) traffic.

Think:

Application-level traffic

Network Load Balancer

Used for network-level traffic.

Think:

Network/transport-level traffic

The important idea:

Google Cloud load balancing is implemented through Google's software-defined network infrastructure rather than requiring you to deploy a physical load-balancer appliance.

15. Autoscaling

Compute Engine can automatically increase or decrease the number of VM instances based on rules.

Example:


Normal traffic
     ↓
3 VMs

Traffic increases
     ↓
Autoscaling
     ↓
5 VMs

Traffic decreases
     ↓
Autoscaling
     ↓
3 VMs

You define the conditions/rules.

Key point

Autoscaling = automatically adjust VM capacity according to demand.

This topic will be covered separately in more detail.

16. Important Compute Engine features

The module will cover:

Machine rightsizing

Choosing an appropriately sized VM.

Example:


Too small → poor performance
Too large → unnecessary cost

Goal:

Right resources + right cost

Startup and shutdown scripts

Scripts that automatically execute when the VM starts or shuts down.

Useful for:

Installing software
Configuring services
Starting applications
Performing cleanup

Metadata

Information/configuration associated with a VM or project.

It can also be used by applications/scripts running on the VM.

Availability policies

Controls related to VM availability and behavior during infrastructure events.

OS patch management

Helps manage operating-system updates and patches.

Pricing and usage discounts

Google Cloud provides different pricing mechanisms/discounts depending on usage and configuration.

17. Cloud TPU

Now there is another important concept:

TPU = Tensor Processing Unit

A TPU is Google's custom-designed hardware for machine learning workloads.

Unlike CPUs and GPUs, TPUs are designed specifically to accelerate certain ML computations.


CPU
↓
General-purpose computing

GPU
↓
Parallel processing / ML

TPU
↓
Specialized ML acceleration

18. Why TPU?

Modern ML models require huge amounts of computation.

A major operation in ML is:

Matrix multiplication

TPUs are designed to perform these kinds of operations efficiently.

Therefore they can provide:

High ML performance
High efficiency
Better energy efficiency for suitable workloads

19. CPU vs GPU vs TPU

Hardware	Main purpose
CPU	General-purpose computing
GPU	Highly parallel workloads, graphics, ML
TPU	Specialized ML workloads

Easy example

Imagine three workers:

CPU:

"I can do almost any type of job."

GPU:

"I can do thousands of similar calculations in parallel."

TPU:

"I'm specially optimized for the mathematical operations common in ML."

20. When are TPUs useful?

TPUs are particularly suitable for:

Large ML models
Long-running training jobs
Large effective batch sizes
ML workloads that benefit from TPU architecture

They aren't automatically the best choice for every ML workload.

⭐ Final Revision Sheet

Remember these points:


Compute Engine
      ↓
Virtual Machines
      ↓
IaaS
      ↓
High flexibility
      ↓
CPU + RAM + Disk + Network + OS

Compute Engine

VM service
IaaS
Supports Linux and Windows
Good for traditional/enterprise applications
Good for on-premises migration
Predefined + custom machine types
Supports autoscaling
Supports load balancing
Supports startup/shutdown scripts
Supports OS patch management

Storage


Standard HDD
→ cheaper
→ capacity-focused

SSD Persistent Disk
→ faster
→ higher IOPS

Local SSD
→ extremely fast
→ low latency
→ temporary/ephemeral

Hardware


CPU → general purpose
GPU → parallel processing
TPU → specialized ML

One-line interview answer

Google Compute Engine is an IaaS service that provides configurable virtual machines with control over CPU, memory, storage, networking, and operating systems, making it suitable for traditional enterprise and on-premises workloads that need high infrastructure flexibility.